1
00:00:12,440 --> 00:00:14,540
Hey everyone and welcome to pie torch.

2
00:00:14,600 --> 00:00:17,600
Deep learning and artificial intelligence.

3
00:00:17,600 --> 00:00:21,400
Now this is one of the most exciting courses I've ever made.

4
00:00:21,440 --> 00:00:26,840
I've been making these deep learning courses for a long time before pi talks was even invented.

5
00:00:26,840 --> 00:00:31,860
So it's been quite a journey for me to see how this field has evolved in the past few years.

6
00:00:31,940 --> 00:00:33,980
And guys it's been moving quick.

7
00:00:34,010 --> 00:00:38,370
The best practices and guidelines you'll learn in this course are the latest and greatest.

8
00:00:38,420 --> 00:00:43,220
This stuff did not even exist when some of my previous deep learning courses came out.

9
00:00:43,220 --> 00:00:47,460
Things like the best way to train your neural network to architecture decisions.

10
00:00:47,570 --> 00:00:51,310
Thanks to the hard work of deep learning researchers all around the world.

11
00:00:51,440 --> 00:00:55,030
I've been able to take what they've learned and bring that information to you.

12
00:00:56,500 --> 00:01:01,960
This course is designed to be a beginner to advanced course so you don't need a lot of math and background

13
00:01:01,960 --> 00:01:07,290
knowledge but of course if you're into that sort of thing I certainly have resources for you.

14
00:01:07,330 --> 00:01:11,530
We're going to start by looking at basic machine learning in the form of a neuron.

15
00:01:11,530 --> 00:01:17,090
The fundamental building block of neural networks from there we are going to jump right into neural

16
00:01:17,090 --> 00:01:17,600
networks.

17
00:01:17,600 --> 00:01:19,010
The thing that started it all.

18
00:01:19,880 --> 00:01:26,090
Luckily with PI talk you don't need a lot of heavy theory unless you wanted the PI talk API already

19
00:01:26,090 --> 00:01:32,150
implements a set of composed all building blocks so you can focus on building cool things rather than

20
00:01:32,150 --> 00:01:35,690
debugging and mathematical equations.

21
00:01:35,760 --> 00:01:40,230
Next we're going to look at convolution on their own networks which are specialized neural networks

22
00:01:40,380 --> 00:01:42,260
for computer vision.

23
00:01:42,300 --> 00:01:46,980
Then we'll look at recurrent neural networks which are specialized neural networks for sequence data

24
00:01:47,010 --> 00:01:50,430
such as Time series text speech and DNA.

25
00:01:50,640 --> 00:01:53,680
We'll even apply our own ends to stock prediction.

26
00:01:53,700 --> 00:01:59,130
This is one of my favorite exercises from this course because it teaches you what 90 percent of other

27
00:01:59,130 --> 00:02:00,330
people are doing wrong.

28
00:02:01,020 --> 00:02:06,450
If you've ever used an LSD GM for stock prediction before or taking a class that promised to teach you

29
00:02:06,750 --> 00:02:13,670
you'll definitely want to watch this because pi talk has a modern and easy to use interface.

30
00:02:13,760 --> 00:02:18,290
It makes building extremely complex things much easier than they used to be.

31
00:02:18,290 --> 00:02:26,770
Here is some example applications first we have Ganz which stands for generative adversarial networks.

32
00:02:26,770 --> 00:02:32,770
These are a new training paradigm which allow you to use deep learning to generate beautiful photo realistic

33
00:02:32,800 --> 00:02:34,060
images.

34
00:02:34,060 --> 00:02:39,460
What's absolutely crazy about this is that the images you are looking at now are people that do not

35
00:02:39,460 --> 00:02:42,870
actually exist but it doesn't stop there.

36
00:02:43,150 --> 00:02:47,860
Deep Learning has been applied in the field of reinforcement learning which excels at tasks that take

37
00:02:47,860 --> 00:02:49,410
multiple steps to complete.

38
00:02:49,420 --> 00:02:51,060
Like playing a video game.

39
00:02:51,370 --> 00:02:55,630
Deep reinforcement learning agents have beat world champions in the strategy game go.

40
00:02:55,690 --> 00:03:02,110
And in modern video games such as CSO and Dota 2 These are things we wouldn't have dreamed of just 10

41
00:03:02,110 --> 00:03:03,430
years ago.

42
00:03:03,520 --> 00:03:07,550
Of course if that's not enough we still have more deep learning powers.

43
00:03:07,630 --> 00:03:14,150
State of the art natural language processing applications such as speech recognition and machine translation.

44
00:03:14,350 --> 00:03:19,180
So whenever you talk to your phone and it figures out that you want to say order a pizza or send in

45
00:03:19,180 --> 00:03:26,260
a line tomorrow for 6 a.m. That's speech recognition using deep learning and action neural machine translation

46
00:03:26,260 --> 00:03:32,080
has significantly enhanced the capabilities of language translation in the recent years with students

47
00:03:32,080 --> 00:03:34,240
in 200 countries around the world.

48
00:03:34,270 --> 00:03:37,050
I've made use of this great technology more than once.

49
00:03:37,300 --> 00:03:40,630
And guys I could go on there's self-driving vehicles.

50
00:03:40,630 --> 00:03:47,380
Object detection facial recognition deep fakes synthesizing speech so you can have a phone conversation

51
00:03:47,380 --> 00:03:48,330
with a robot.

52
00:03:49,580 --> 00:03:53,790
So if you want to learn how to use the world's fastest rising Deep Learning library.

53
00:03:53,840 --> 00:03:55,940
This course is for you.

54
00:03:55,940 --> 00:03:58,130
Thanks for listening and I'll see you in the next lecture.